MiniMax H3 reference to video · Auto Prompt + upscale
MiniMax H3 reference-to-video with auto prompting and a 2x upscale. Keep your AI character consistent in sharp AI video with sound.
MiniMax H3Image, Video, Text → Video, AudioAdvanced
MiniMax H3 is an open AI video model that generates video with sound in one pass: voices, sound effects and ambience come out already synced to the picture. Give it references of your character, let AI write the prompt, and get a sharper 2x upscaled clip at the end.
Great for:
High-quality, consistent AI influencer videos
Recurring cartoon characters in polished clips
Longer 7-second scenes
How it works
Add your OpenRouter API key.
Add your references and type a short idea.
Run it. You get the base video and a 2x upscaled version, both with audio.
Make it yours with LoRAs
This workflow supports LoRAs. Grab character, style, anime or cartoon LoRAs from CivitAI or Hugging Face, drop them into the LoRA loader, and keep your AI character looking the same in every shot.
Recommended settings6
GPU
H100
H200
RTX PRO 6000
RTX 5090
Sampler
euler
res_multistep
Scheduler
simple
Turbo LoRA
LightX2V 8-step turbo LoRA
Steps
8 with the turbo LoRA at full strength
10-12 with the turbo LoRA at lower strength for better quality
Hugging Face